Multipopulation-based multi-level parallel enhanced Jaya algorithms

Abstract

To solve optimization problems, in the field of engineering optimization, an optimal value of a specific function must be found, in a limited time, within a constrained or unconstrained domain. Metaheuristic methods are useful for a wide range of scientific and engineering applications, which accelerate being able to achieve optimal or near-optimal solutions. The metaheuristic method called Jaya has generated growing interest because of its simplicity and efficiency. We present Jaya-based parallel algorithms to efficiently exploit cluster computing platforms (heterogeneous memory platforms). We propose a multi-level parallel algorithm, in which, to exploit distributed-memory architectures (or multiprocessors), the outermost layer of the Jaya algorithm is parallelized. Moreover, in internal layers, we exploit shared-memory architectures (or multicores) by adding two more levels of parallelization. This two-level internal parallel algorithm is based on both a multipopulation structure and an improved heuristic search path relative to the search path of the sequential algorithm. The multi-level parallel algorithm obtains average efficiency values of 84% using up to 120 and 135 processes, and slightly accelerates the convergence with respect to the sequential Jaya algorithm.